Study Verifiable Knowledge at Scale
As superintelligence (SI) becomes more deeply involved in industrial production and organizational decisions, we need to understand when systems deserve our trust, how safety requirements can be verified, and how different goals and incentives shape behavior when people, SI, and organizations make decisions together.
We want practical problems to inform theoretical research, and methods and insights from theory to be tested in real settings.
SI safety and industrial practice
When SI enters a production workflow, safety must be addressed in the specific system, environment, and operations involved. What is the system allowed to do? Which behaviors must be constrained? How should it respond to uncertainty or tasks beyond its capabilities?
We study how industrial safety requirements can become precise specifications, verifiable properties, and safeguards that can be put into practice. We combine these with interpretability, system evaluation, and monitoring during operation to understand SI behavior and its effects.
We also study the boundaries of safety guarantees: which assumptions a verification depends on, which situations it covers, and whether those guarantees still hold when the environment changes.
Mathematics, formalization, and verification
Mathematics helps us express problems precisely. Formal methods allow reasoning and system behavior to be checked rigorously.
We explore proof tools such as Lean and SI for mathematics: how they can support mathematical discovery, map the dependencies of arguments, and connect formal proofs to understandable mathematical ideas. Our interests extend from pure mathematics, including geometric analysis, to software verification and formal models of industrial systems.
Connecting theory and practice requires both sides to clarify the problem together: which practical constraints belong in a model, which properties are worth proving, and what decisions a proof can support in practice.
Games, incentives, and cooperation
SI behavior is also shaped by the organizations and cooperative arrangements around it. Participants have different information, goals, and resources; rules and incentives change their choices.
We study strategic interactions among people, SI, and organizations: how to encourage truthful information sharing, how to distribute the costs and responsibilities of verification, and how to prevent local gains from motivating unsafe behavior.
Through game theory and mechanism design, we explore the conditions for reliable cooperation, and the limits of these mechanisms when information is incomplete, goals conflict, or participants act strategically.
Open science and shared knowledge
Research itself faces questions of cooperation and incentives: what work receives funding, who checks the results, how contributions are recorded, and how others can continue using the knowledge produced.
We see decentralized science (DeSci) as one path for exploring these questions. We study how cryptographic tools, funding mechanisms, and knowledge markets can support open research, preserve provenance, and give appropriate recognition to different kinds of contribution.
We ask whether these mechanisms help communities sustain knowledge that can be checked, reused, and built upon. We also examine the speculation, biases, and new costs of cooperation they may introduce.
Learning through shared research
We learn through shared reading, discussion of practical problems, formalizing arguments, and building small tools. We want theoretical researchers and industry practitioners to work together on concrete questions, starting with explicit assumptions and gradually developing models, proofs, and experiments that can be tested.
Explore our research. Whether you bring a theoretical question, a technical method, or an unresolved problem from production, you are welcome to study it with us.